what-was-that

You are mid-conversation with an AI assistant and it drops a string of RAG, LoRA, and MCP terms that leave you frozen. that is exactly the moment what-was-that is built for. A cute little octopus in the bottom-right corner watches the AI reply in real time; when it spots a term from the built-in 255-entry glossary, a red badge and a gentle tooltip appear. One click and the concept is explained in plain language with everyday analogies, entirely offline and at zero token cost.

Want to dig deeper? Tap “go one step further” and the AI walks through a short explanation, a rephrased version, a longer breakdown, and a bottom-line analogy in successive rounds. You can also highlight any short word (up to 10 characters) right in the chat area and an explanation card pops up beside the selection. The octopus itself has four moods, is draggable, and adapts to light or dark themes using lightweight transparent SVGs.

The whole package is just 212 KB compressed, has zero runtime dependencies, and calls only the DSH runtime API. Privacy-wise it never sends the full reply back to the model; only the selected word plus 80 characters of surrounding context, cached locally. It registers no model tools and never reads chain-of-thought, quietly doing one job: making jargon feel human. Ideal for programming newcomers who get shut out by buzzwords, non-technical professionals who read AI output daily, and anyone who wants an instant plain-English translation right in the conversation.